This article can also be found in the Premium Editorial Download "Storage magazine: Distance: the new mantra for disaster recovery."
Download it now to read this article plus other related content.
Distributed computing model
Some companies are moving away from expensive, proprietary systems to low-cost Linux clusters. Minneapolis, MN-based Sistina is a good example of this. Its Global File System (GFS) is an outgrowth of a project at the University of Minnesota. According to Joaquin Ruiz, VP of product management, "With Sistina, you can just add bricks of compute power without having to do a forklift upgrade."
Sistina and other companies are hoping to ride the conversion of large midrange and mainframe applications to Linux, where a shared file system can be used between storage and low-cost Linux servers to help connect a database, scientific or custom application cluster. Shared file systems allow you to add storage or servers as capacity is needed, instead of doing a big upgrade of a central server.
PolyServe, Beaverton, OR, is another company with a build-as-you-need philosophy. For example, Steve Norall, director of product marketing at PolyServe, says, "Matrix Server is targeted at Global 2000 data centers that are focused on building highly available, scalable Intel-based server farms." According to Norall, the product is a fully symmetric cluster file system with a lock and metadata manager.
IBM also offers a single-platform clustered file system, its General Parallel File System (GPFS), designed and used primarily for parallel computing. IBM's Tevis says, "GPFS is focused on a different set of applications than StorageTank. GPFS is ideal for environments like scientific computing where a clustered file system with high performance for parallel access is desired." (See "File sharing product roundup")Clifford Baeseman, Linux administrator at Greenheck Fan Corporation in Schofield, Wisconsin, a manufacturer of ventilation equipment, is using Sistina's GFS on a 1.5TB SAN. Greenheck is running several clustered file systems--one with Sistina GFS, one running Oracle RAC with raw device--and one running Oracle Cluster File System (OCFS), as well as an AlphaVMS cluster they've been running since 1986. "We're running Linux Terminal Server Project, serving X-windows desktops from a single machine out to our manufacturing floor. One machine services 70 desktops," Baeseman says, adding, "If that goes down, all manufacturing stops. Our need for clustering drove us to migrate to a SAN." One of the primary reasons Baeseman is using file sharing software is to ensure the high availability of his servers.
Baeseman looked at several different clustered file systems, including PolyServe, which he didn't pick because "we didn't like the stability yet." He's now running Sistina's GFS, which passed his "brutality testing," which consists of scripts which access the same files from multiple systems and check for data corruption. However, Baeseman actually prefers open-source (GPL) solutions, so he can switch over to OCFS as it matures.
Greenheck has been slowly converting their systems over to Linux-based clusters attached to their SAN. Schreiber explains, "The goal of this project is to get the same stability as Digital Alpha VMS." He's been happy with the solution, saying that the Linux clusters are bringing "significantly lower cost than traditional mainframe product suites."
Despite the promised advantages of shared file systems, users have been slow to adopt the technology. IBM, although continuing to sell and support SANergy, indicates the future lies with its StorageTank products. Similarly, Veritas has refocused its cluster file system offerings to target a few specific markets, such as the Oracle RAC.
When asked how EMC's HighRoad solution is selling, Ross admits "adoption is modest," adding, "it's a new technology and people need to understand what it's used for." Ross says the high cost of SAN equipment is slowing the adoption rate of all file sharing products. "You need a SAN to get the benefits," he says.
Werner Zurcher, a product manager at Veritas agrees that the expense of a SAN is currently a significant barrier to adoption. He says, "Practically speaking, cluster file systems require a SAN, and currently there are a limited set of customers who are willing to pay extra money for the SAN infrastructure needed to connect multiple systems to some shared disks." He explains, "The reason clustered file systems have been successful in some vertical markets--such as geophysics and multimedia--is because the application files in those market segments tend to be large enough that going to a clustered file system makes a big performance difference. File sharing via Ethernet provides reasonable I/O performance for small files, but it does not scale very well for large files."
IBM's Tevis is bullish on the future of shared file system technology. He says clustered file systems will continue to improve and win market acceptance. As he put it: "Clustered file systems are not going away, and the industry is heading more and more in this direction."
This was first published in May 2003